Claude Code Provider

Use Repowise with your Claude subscription via the claude_cli LLM provider. No API key — the Claude Code CLI holds the auth.

Table of contents

  1. Quick setup
  2. Two directions, one name
  3. Choosing a model
  4. Cost
  5. Reasoning
  6. Embeddings are separate
  7. Troubleshooting

Quick setup 

# 1. Install Claude Code, then authenticate once
claude auth login

# 2. Point Repowise at it
repowise init --provider claude_cli --yes

Any plan that can run claude -p works — Pro, Max, Team or Enterprise.

To use it on an index you already have:

REPOWISE_PROVIDER=claude_cli repowise generate --unwritten

Or persist it in .repowise/config.yaml:

provider: claude_cli
model: claude_cli/claude-haiku-5-5

Two directions, one name

This page is about Repowise calling Claude Code to write your wiki. The opposite direction — Claude Code calling Repowise’s MCP tools to answer questions about your codebase — is the claude-code agent target, set up with repowise agents add --target=claude-code. They are independent; enabling one does not enable the other.

Choosing a model

claude_cli/claude-haiku-5-5 is the default, matching the anthropic provider.

repowise init --provider claude_cli --model claude_cli/claude-sonnet-4-6

The claude_cli/ prefix is optional on input:

repowise init --provider claude_cli --model claude-opus-4-6
Model Notes
claude-haiku-5-5 Fastest, and ample for doc pages. Default.
claude-sonnet-4-6 Better prose, slower.
claude-opus-4-6 Highest quality, heaviest on subscription limits.

Cost

claude_cli/* is priced at $0.00 in cost estimates and the cost history, because a subscription is not per-token API spend. The CLI’s own reported cost is kept under usage.reported_cost_usd for auditing.

The token volume is real and reported: Claude Code splits a prompt across input_tokens (the uncached remainder), cache_creation_input_tokens (a cache write) and cache_read_input_tokens (a cache read), and only the first is usually more than a couple of tokens for a page prompt. Repowise records the sum as the run’s input tokens, so repowise status and repowise costs show the prompt the model actually read rather than a figure near zero.

What it does consume is your subscription rate limits, the same ones your interactive Claude Code sessions use. Budget roughly 9k output tokens and a few minutes per page. A full wiki on a large repo is a real chunk of usage, so scope runs with --path rather than reaching for --all:

repowise generate --path src/api

Concurrency defaults to 4 processes. The variable below overrides it in either direction – lower it if you are hitting limits, and note that raising it past 4 is allowed but is the quickest way to hit them:

REPOWISE_CLAUDE_CLI_CONCURRENCY=2 repowise generate --unwritten

Reasoning

Claude Code’s --effort flag maps directly to Repowise’s low, medium, high, xhigh, and max reasoning modes. auto leaves the flag unset and uses the CLI default. The unsupported off, none, and minimal modes log a warning and proceed with the default rather than failing the run.

Embeddings are separate

Anthropic has no embeddings API, so claude_cli cannot serve as an embedder. If you want semantic search rather than keyword-only, pick an embedder separately:

repowise init --provider claude_cli --embedder ollama   # local, no key

Otherwise the index keeps embedder: mock and search stays lexical.

Troubleshooting

Claude Code CLI not found — claude is not on PATH. Install it from claude.com/claude-code.

Authentication errors on the first page — run claude auth login, then check with claude auth status, which exits 0 when you are signed in. That is the check repowise init uses to show the provider as ready.

Rate-limit failures partway through a run — lower REPOWISE_CLAUDE_CLI_CONCURRENCY, or scope the run with --path. Completed pages are kept, so re-running picks up where it stopped with --unwritten.